Introduction

This document describes the re-estimation of the VE Multimodal Module (VE-MM) using the 2017 National Household Travel Survey (2017 NHTS). The original version of the VE-MM was estimated using the 2009 NHTS.

The models described in this memo include:

  • regression models used to predict household average annual daily vehicle miles traveled (AADVMT)
  • models used to predict transit, bicycle, and walking trips and person miles traveled (note that these are not included in this draft)

Estimation Data

Data Sources

Data used to estimate the VE-MM models came from several sources. The 2017 NHTS confidential data were obtained from ORNL via FHWA. The data were provided with Census Block Group which allowed data from the EPA Smart Location Database, version 3 (SLD) to be connected to the household. Additional spatial information at the metropolitan area level were added to the data including transit revenue miles from the National Transit Database and freeway lane miles from FHWA published data.

AADVMT and DVMT Data

The AADVMT model is estimated using the total annual miles driven by vehicles owned by the NHTS households divided by 365, instead of estimating a daily VMT model based on the reported travel during the survey day. The distributions of those travel amounts differs in the data.

In particular there are a lot of “zero days” in the survey data for the survey day. There are also more very long days in the survey data with more than 200 miles of travel reported. With the averaging over the year that takes place by using AADVMT, a lot of the day to day variability within a household’s travel is removed. The final chart compares the none-zero day distribution and other than the extremes at the low end and high end of the distribution, the distributions are reasonably comparable.

Distribution of AADVMT per Household in the 2017 NHTS

Figure 1: Distribution of AADVMT per Household in the 2017 NHTS

Distribution of Survey Day DVMT per Household in the 2017 NHTS

Figure 2: Distribution of Survey Day DVMT per Household in the 2017 NHTS

Comparison between AADVMT and Survey Day DVMT per Household in the 2017 NHTS

Figure 3: Comparison between AADVMT and Survey Day DVMT per Household in the 2017 NHTS

AADVMT Outlier Filtering

The households with the highest 1% of AADVMT were excluded from the estimation dataset. The exclusion of outliers has the effect of reducing the average AADVMT per household. The table below shows the impact on the average AADVMT per household for metro and non-metro households. The charts shows the change in the binned distribution of households for metro and non-metro households.

All of the analysis in the remainder of this document is based on the 2017 NHTS data with outliers removed, consistent with the set of households used for estimation.

Outlier Status Metro HH Non-Metro HH Metro Avg. AADVMT Non-Metro Avg. AADVMT
Included 71740 56646 47.54 61.01
Outlier 617 693 359.50 293.96
Total 72357 57339 50.85 64.96
Impacts of Outlier Filtering on the AADVMT per Household in the 2017 NHTS

Figure 4: Impacts of Outlier Filtering on the AADVMT per Household in the 2017 NHTS

Population Density (D1B)

One of the important explanatory variables for household AADVMT is the population density of the neighborhood in which the household resides. The 2017 NHTS data were linked to the SLD and the D1B variable, which contains the Census Block Group population density in units of persons per acre. The ranges of density varies significantly in the sample of NHTS households. As is expected, most non-metro households live in low density areas while the range of densities for metropolitan households is generally higher but covers several orders of magnitude. The following box plot, plotted on a log scale shows the distribution of households by population density.

Distribution of Households in the 2017 NHTS by Population Density

Figure 5: Distribution of Households in the 2017 NHTS by Population Density

The following charts show the relationship between AADVMT in low, medium, and high density areas, split by metro and non-metro households in low and medium density areas, and showing just metropolitan households in the high density chart. The trends in the charts show a clear relationship between density and AADVMT: as density increases, AADVMT tends to fall. There are some anomalous values, particular for non-metropolitan households, due to small sample sizes.

2017 NHTS AADVMT by Density for Households in Low Density Neighborhoods

Figure 6: 2017 NHTS AADVMT by Density for Households in Low Density Neighborhoods

2017 NHTS AADVMT by Density for Households in Medium Density Neighborhoods

Figure 7: 2017 NHTS AADVMT by Density for Households in Medium Density Neighborhoods

2017 NHTS AADVMT by Density for Households in High Density Neighborhoods

Figure 8: 2017 NHTS AADVMT by Density for Households in High Density Neighborhoods

Household Income

Household income has a positive impact on the amount of travel a household makes, with higher income households traveling more. The first chart shows the distribution of income levels amongst the households in the 2017 NHTS sample.

Just over 4,000 (3%) of the households in the sample were missing a response to the household income question. For the purpose of this analysis they have been recoded with an income in the median income category, which has a midpoint of $62,500.

The second chart shows the relationship between AADVMT and household income and shows shows a clear postiive trend for both metropolitan and non-metropolitan households.

Distribution of Households in the 2017 NHTS by Household Income

Figure 9: Distribution of Households in the 2017 NHTS by Household Income

2017 NHTS AADVMT by Household Income

Figure 10: 2017 NHTS AADVMT by Household Income

Household Size

Household size has a positive impact on the amount of travel a household makes, with larger households traveling more. The first chart shows the distribution of household size amongst the households in the 2017 NHTS sample. The second charts shows the relationship between AADVMT and household size and shows shows a clear postiive trend for both metropolitan and non-metropolitan households.

Distribution of Households in the 2017 NHTS by Household Size

Figure 11: Distribution of Households in the 2017 NHTS by Household Size

2017 NHTS AADVMT by Household Size

Figure 12: 2017 NHTS AADVMT by Household Size

Household life cycle stage

The household’s life cycle stage, i.e., whether the household is a single person, a couple without children, a couple with children, or older “empty nesters” influences the amount of travel a household makes. Multi person household make more travel and adding children to the household causes a further moderate increase in the amount of travel. The first chart shows the distribution of life cycle stage amongst the households in the 2017 NHTS sample. The second charts shows the relationship between AADVMT and household life cycle stage for both metropolitan and non-metropolitan households.

Distribution of Households in the 2017 NHTS by Household Life Cycle Stage

Figure 13: Distribution of Households in the 2017 NHTS by Household Life Cycle Stage

2017 NHTS AADVMT by Household Life Cycle Stage

Figure 14: 2017 NHTS AADVMT by Household Life Cycle Stage

Number of Workers

The number of workers in the household positively impacts the amount of travel a household makes. With each additional worker in the household, the household make more travel. The first chart shows the distribution of number of workers per households amongst the households in the 2017 NHTS sample. The second charts shows the relationship between AADVMT and number of workers per household for both metropolitan and non-metropolitan households.

Distribution of Households in the 2017 NHTS by Number of Workers in the Household

Figure 15: Distribution of Households in the 2017 NHTS by Number of Workers in the Household

2017 NHTS AADVMT by Number of Workers in the Household

Figure 16: 2017 NHTS AADVMT by Number of Workers in the Household

Number of Drivers

The number of drivers in the household positively impacts the amount of travel a household makes. With each additional driver in the household, the household make more travel. The first chart shows the distribution of number of drivers per households amongst the households in the 2017 NHTS sample. The second charts shows the relationship between AADVMT and number of drivers per household for both metropolitan and non-metropolitan households.

Distribution of Households in the 2017 NHTS by Number of Drivers in the Household

Figure 17: Distribution of Households in the 2017 NHTS by Number of Drivers in the Household

2017 NHTS AADVMT by Number of Drivers in the Household

Figure 18: 2017 NHTS AADVMT by Number of Drivers in the Household

Oregon Households

The 2017 NHTS sample includes some households in Oregon. Oregon was not an “add on” state and therefore has sample from just the all-US sample, a total of 385 households. Average AADVMt per household in metro areas is slightly lower in Oregon than the average for the rest of the US, and is significantly lower for non-metro households.

State Metro HH Non-Metro HH Metro Avg. AADVMT Non-Metro Avg. AADVMT
Oregon 236 148 47.32 55.97
Other State 71504 56498 47.54 61.09

The income distribution for the Oregon households is reasonably similar to that for the other states, with a notable difference being a low number of high income households in non-metro areas. The positive effect of income of AADVMT is clear, although it is negligible across the upper categories in metro areas. The data for non-metro areas is less reliable for the high income categories due to small sample sizes.

Oregon and Other State Household Income Distribution

Figure 19: Oregon and Other State Household Income Distribution

Oregon and Other State AADVMT by Household Income

Figure 20: Oregon and Other State AADVMT by Household Income

Model Estimation

2017 NHTS AADVMT Model

The following table shows the AADVMT model estimated using the 2017 NHTS for Metro areas and Non-Metro area.

Regression model summary for 2017 AADVMT Model for Metro Areas and Non-Metro Areas
Dependent variable:
AADVMT
NONMETRO METRO
(1) (2)
Drivers 0.929*** (0.012) 0.995*** (0.008)
HhSize 0.105*** (0.010)
Workers 0.234*** (0.009) 0.151*** (0.008)
CENSUS_RNE -0.069*** (0.019) -0.085*** (0.016)
CENSUS_RS 0.095*** (0.014) 0.012 (0.014)
CENSUS_RW -0.212*** (0.017) -0.119*** (0.015)
FwyLaneMiPC 49.622*** (17.024)
LogIncomeK 0.351*** (0.007) 0.190*** (0.006)
Age0to14 -0.003 (0.012) 0.090*** (0.009)
Age65Plus -0.072*** (0.011) -0.096*** (0.010)
log1p(VehPerDriver) 4.105*** (0.025) 4.262*** (0.021)
LifeCycleCouple w/o children -0.033 (0.021) -0.059*** (0.016)
LifeCycleEmpty Nester -0.267*** (0.026) -0.456*** (0.020)
LifeCycleSingle -0.211*** (0.028) -0.413*** (0.018)
D1B -0.019*** (0.003) -0.002*** (0.0002)
D2A_EPHHM -0.298*** (0.029)
D1B:D2A_EPHHM 0.023*** (0.005)
D2A_WRKEMP -0.0002 (0.0002)
D3bpo4 -0.001*** (0.0001)
TranRevMiPC:D4c -0.056*** (0.004)
Constant -0.802*** (0.045) -0.320*** (0.031)
Observations 56,551 71,740
R2 0.654 0.700
Adjusted R2 0.654 0.700
Residual Std. Error 1.182 (df = 56534) 1.384 (df = 71722)
F Statistic 6,671.994*** (df = 16; 56534) 9,861.924*** (df = 17; 71722)
Note: p<0.1; p<0.05; p<0.01

The following tables compare the AADVMT models estimated using the 2009 and 2017 NHTS.

This first comparison is between 2009 and 2017 AADVMT Models for Metro areas.

VarName NHTS2009 NHTS2017 Ratio
(Intercept) -1.333 -0.320 0.240
Age0to14 0.107 0.090 0.840
Age65Plus -0.075 -0.096 1.286
CENSUS_RNE -0.109 -0.085 0.777
CENSUS_RS 0.051 0.012 0.227
CENSUS_RW -0.092 -0.119 1.289
D1B -0.003 -0.002 0.690
D2A_WRKEMP 0.000 0.000 0.914
D3bpo4 -0.001 -0.001 0.892
Drivers 0.705 0.995 1.412
FwyLaneMiPC 101.341 49.622 0.490
LifeCycleCouple w/o children -0.036 -0.059 1.611
LifeCycleEmpty Nester -0.256 -0.456 1.778
LifeCycleSingle -0.234 -0.413 1.767
LogIncome 0.268 0.000 0.000
LogIncomeK 0.000 0.190 Inf
TranRevMiPC:D4c -0.020 -0.056 2.821
Workers 0.186 0.151 0.811
log1p(VehPerDriver) 1.794 4.262 2.376

This second comparison is between 2009 and 2017 AADVMT Models for Non-Metro areas.

VarName NHTS2009 NHTS2017 Ratio
(Intercept) -1.416 -0.802 0.566
Age0to14 0.102 -0.003 -0.031
Age65Plus -0.077 -0.072 0.932
CENSUS_RNE -0.112 -0.069 0.617
CENSUS_RS 0.058 0.095 1.632
CENSUS_RW -0.176 -0.212 1.206
D1B -0.008 -0.019 2.325
D1B:D2A_EPHHM -0.027 0.023 -0.867
D2A_EPHHM -0.084 -0.298 3.524
Drivers 0.744 0.929 1.249
HhSize 0.017 0.105 6.231
LifeCycleCouple w/o children -0.013 -0.033 2.421
LifeCycleEmpty Nester -0.208 -0.267 1.281
LifeCycleSingle -0.216 -0.211 0.975
LogIncome 0.288 0.000 0.000
LogIncomeK 0.000 0.351 Inf
Workers 0.177 0.234 1.322
log1p(VehPerDriver) 1.852 4.105 2.216

2017 NHTS AADVMT Model Adjustment

One aspect of the NHTS AADVMT data that is hard to capture, even with a power transform adjusted linear model, is the variability in the data due to unobserved household travel characteristics. Some households just travel more or less than other similar households with similar income, transportation access, and neighborhood characteristics.

In order to capture this dispersion, a random variable factor has been drawn for each household to factor the predicted AADVMT to simulate household to household variation. This random variable is drawn from a left skewed normal distribution to allow for some households to have lower AADVMT, to overall achieve a slight increase in AADVMT to account for systematic under prediction of the mean AADVMT in the sample, and to produce a longer tail of households with a higher AADVMT.

Skewed Normal Distribution

Figure 21: Skewed Normal Distribution

Model Prediction Testing

Households by Metro and Non-Metro

Metro or Non-Metro Weighted HH NHTS2017 AADVMT/HH Model AADVMT/HH Model Sim AADVMT/HH Ratio Model/NHTS2017 Ratio Model Sim/NHTS2017
metro 81842.28 47.54 43.17 47.61 0.91 1
non_metro 46121.44 61.04 56.58 61.06 0.93 1
Scatterplot of Model Prediction vs. 2017 NHTS Data

Figure 22: Scatterplot of Model Prediction vs. 2017 NHTS Data

Scatterplot of Model Prediction (Simulated) vs. 2017 NHTS Data

(#fig:scatter-aadvmt-model-prediction_rnd)Scatterplot of Model Prediction (Simulated) vs. 2017 NHTS Data

Households by DVMT Bins, Model Prediction vs. 2017 NHTS Data

Figure 23: Households by DVMT Bins, Model Prediction vs. 2017 NHTS Data

Difference in Households in DVMT Bins (Model Prediction - 2017 NHTS Data)

Figure 24: Difference in Households in DVMT Bins (Model Prediction - 2017 NHTS Data)

Difference in Households in DVMT Bins (Model Prediction (Simulated) - 2017 NHTS Data)

Figure 25: Difference in Households in DVMT Bins (Model Prediction (Simulated) - 2017 NHTS Data)

Difference in Households in DVMT Bins (Model Prediction and Simulated - 2017 NHTS Data)

Figure 26: Difference in Households in DVMT Bins (Model Prediction and Simulated - 2017 NHTS Data)

Households by Income

HH Income Metro WgtHH Non-Metro WgtHH Metro Obs AADVMT/HH Non-Metro Obs AADVMT/HH Metro Pred AADVMT/HH Non-Metro Pred AADVMT/HH Metro Sim AADVMT/HH Non-Metro Sim AADVMT/HH Metro Ratio Pred/Obs Non-Metro Ratio Pred/Obs Metro Ratio Sim/Obs Non-Metro Ratio Sim/Obs
5000 6086.08 3265.04 18.53 20.41 15.01 16.83 16.50 18.54 0.81 0.82 0.89 0.91
12500 4376.62 3136.07 21.29 29.89 19.51 25.93 21.59 28.02 0.92 0.87 1.01 0.94
19999 7380.65 4786.65 29.94 40.09 26.98 35.28 29.69 37.92 0.90 0.88 0.99 0.95
30000 7568.41 4899.85 37.60 48.02 33.40 43.04 37.00 46.02 0.89 0.90 0.98 0.96
42499 9554.17 5807.06 44.58 57.59 39.06 53.08 42.90 57.24 0.88 0.92 0.96 0.99
62500 15152.85 9288.25 48.06 65.62 43.50 61.75 48.03 66.72 0.91 0.94 1.00 1.02
87500 9827.42 5408.12 56.70 79.17 51.13 74.24 56.23 79.49 0.90 0.94 0.99 1.00
112500 7638.21 4000.40 64.35 88.30 59.21 79.73 65.41 86.15 0.92 0.90 1.02 0.98
137500 4486.90 2099.04 68.08 88.35 62.07 85.63 68.27 92.82 0.91 0.97 1.00 1.05
174999 4683.85 1748.72 68.91 91.53 64.81 87.45 71.83 95.68 0.94 0.96 1.04 1.05
249999 5087.13 1682.24 68.33 93.31 65.93 94.16 72.71 102.33 0.96 1.01 1.06 1.10
Households by Income Group, Model Prediction vs. 2017 NHTS Data

Figure 27: Households by Income Group, Model Prediction vs. 2017 NHTS Data

Scatterplot of Model Prediction vs. 2017 NHTS Data by Income

Figure 28: Scatterplot of Model Prediction vs. 2017 NHTS Data by Income

Households by Density

D1B Group Metro WgtHH Non-Metro WgtHH Metro Obs AADVMT/HH Non-Metro Obs AADVMT/HH Metro Pred AADVMT/HH Non-Metro Pred AADVMT/HH Metro Sim AADVMT/HH Non-Metro Sim AADVMT/HH Metro Ratio Pred/Obs Non-Metro Ratio Pred/Obs Metro Ratio Sim/Obs Non-Metro Ratio Sim/Obs
0 50657.88 44851.81 53.35 61.42 48.59 56.92 53.60 61.44 0.91 0.93 1.00 1.00
10 17457.51 1103.06 44.74 48.65 40.84 45.20 45.07 48.04 0.91 0.93 1.01 0.99
20 4807.56 109.48 39.94 48.37 35.23 39.52 38.80 44.01 0.88 0.82 0.97 0.91
30 2442.79 33.11 33.59 36.09 28.41 52.89 31.83 66.00 0.85 1.47 0.95 1.83
40 1380.94 23.97 30.18 13.51 27.10 24.34 29.28 26.51 0.90 1.80 0.97 1.96
50 985.24 0.00 23.74 0.00 22.24 0.00 24.27 0.00 0.94 0.00 1.02 0.00
60 649.40 0.00 22.28 0.00 22.41 0.00 24.29 0.00 1.01 0.00 1.09 0.00
70 445.27 0.00 20.13 0.00 17.65 0.00 19.43 0.00 0.88 0.00 0.97 0.00
80 386.24 0.00 20.74 0.00 18.31 0.00 19.49 0.00 0.88 0.00 0.94 0.00
90 344.47 0.00 17.75 0.00 17.85 0.00 18.56 0.00 1.01 0.00 1.05 0.00
100 1539.17 0.00 14.92 0.00 12.76 0.00 14.26 0.00 0.86 0.00 0.96 0.00
200 745.81 0.00 9.43 0.00 6.90 0.00 7.51 0.00 0.73 0.00 0.80 0.00
Households by Density, Model Prediction vs. 2017 NHTS Data

Figure 29: Households by Density, Model Prediction vs. 2017 NHTS Data

Scatterplot of Model Prediction vs. 2017 NHTS Data by Density

Figure 30: Scatterplot of Model Prediction vs. 2017 NHTS Data by Density

Households by Household Size

HH Size Metro WgtHH Non-Metro WgtHH Metro Obs AADVMT/HH Non-Metro Obs AADVMT/HH Metro Pred AADVMT/HH Non-Metro Pred AADVMT/HH Metro Sim AADVMT/HH Non-Metro Sim AADVMT/HH Metro Ratio Pred/Obs Non-Metro Ratio Pred/Obs Metro Ratio Sim/Obs Non-Metro Ratio Sim/Obs
1 24826.85 11185.76 22.38 28.57 20.53 26.82 22.69 28.89 0.92 0.94 1.01 1.01
2 26637.43 16940.21 47.44 59.50 41.94 53.59 46.23 57.72 0.88 0.90 0.97 0.97
3 12614.19 7323.02 61.62 78.82 55.40 71.32 60.89 76.42 0.90 0.90 0.99 0.97
4 11550.85 6568.33 70.45 84.68 65.36 80.01 72.34 87.58 0.93 0.94 1.03 1.03
5 4232.36 2540.60 75.39 85.51 71.52 84.55 79.14 92.21 0.95 0.99 1.05 1.08
6 1980.60 1563.52 81.19 87.61 75.69 88.87 82.39 93.49 0.93 1.01 1.01 1.07
Households by Size, Model Prediction vs. 2017 NHTS Data

Figure 31: Households by Size, Model Prediction vs. 2017 NHTS Data

Scatterplot of Model Prediction vs. 2017 NHTS Data by Household Size

Figure 32: Scatterplot of Model Prediction vs. 2017 NHTS Data by Household Size

Households by Number of Workers

Workers Metro WgtHH Non-Metro WgtHH Metro Obs AADVMT/HH Non-Metro Obs AADVMT/HH Metro Pred AADVMT/HH Non-Metro Pred AADVMT/HH Metro Sim AADVMT/HH Non-Metro Sim AADVMT/HH Metro Ratio Pred/Obs Non-Metro Ratio Pred/Obs Metro Ratio Sim/Obs Non-Metro Ratio Sim/Obs
0 20288.99 14255.02 23.26 34.24 21.50 31.10 23.69 33.72 0.92 0.91 1.02 0.99
1 32134.24 15903.13 41.30 56.57 36.52 50.97 40.28 54.87 0.88 0.90 0.98 0.97
2 24189.03 13063.12 66.00 83.73 58.99 77.08 64.87 83.02 0.89 0.92 0.98 0.99
3 4088.31 2383.94 90.39 111.44 88.02 113.20 97.34 122.64 0.97 1.02 1.08 1.10
4 956.45 458.84 106.01 130.78 114.64 148.19 130.25 161.17 1.08 1.13 1.23 1.23
5 185.26 57.39 129.07 141.05 147.17 184.89 162.25 210.10 1.14 1.31 1.26 1.49
Households by Workers, Model Prediction vs. 2017 NHTS Data

Figure 33: Households by Workers, Model Prediction vs. 2017 NHTS Data

Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Workers

Figure 34: Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Workers

Households by Number of Drivers

Drivers Metro WgtHH Non-Metro WgtHH Metro Obs AADVMT/HH Non-Metro Obs AADVMT/HH Metro Pred AADVMT/HH Non-Metro Pred AADVMT/HH Metro Sim AADVMT/HH Non-Metro Sim AADVMT/HH Metro Ratio Pred/Obs Non-Metro Ratio Pred/Obs Metro Ratio Sim/Obs Non-Metro Ratio Sim/Obs
0 6384.30 1775.83 0.16 0.79 0.01 0.09 0.01 0.10 0.08 0.12 0.09 0.12
1 29508.16 13948.61 28.85 33.85 26.90 32.31 29.76 34.86 0.93 0.95 1.03 1.03
2 35915.65 23736.80 58.82 69.71 51.45 61.90 56.56 67.00 0.87 0.89 0.96 0.96
3 7266.06 5100.45 84.46 97.00 78.05 92.47 86.28 98.84 0.92 0.95 1.02 1.02
4 2296.87 1276.29 108.70 120.81 110.80 133.56 124.26 146.45 1.02 1.11 1.14 1.21
5 471.24 283.45 132.09 135.04 148.30 166.05 158.42 171.51 1.12 1.23 1.20 1.27
Households by Drivers, Model Prediction vs. 2017 NHTS Data

Figure 35: Households by Drivers, Model Prediction vs. 2017 NHTS Data

Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Drivers

Figure 36: Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Drivers

Households by Number of Vehicles

Vehicles Metro WgtHH Non-Metro WgtHH Metro Obs AADVMT/HH Non-Metro Obs AADVMT/HH Metro Pred AADVMT/HH Non-Metro Pred AADVMT/HH Metro Sim AADVMT/HH Non-Metro Sim AADVMT/HH Metro Ratio Pred/Obs Non-Metro Ratio Pred/Obs Metro Ratio Sim/Obs Non-Metro Ratio Sim/Obs
0 9461.80 2451.54 0.00 0.00 0.81 1.44 0.88 1.49 Inf Inf Inf Inf
1 30836.06 13477.84 29.24 30.42 25.27 25.88 27.91 27.96 0.86 0.85 0.95 0.92
2 27673.26 17077.66 61.01 63.28 54.00 57.67 59.51 62.23 0.89 0.91 0.98 0.98
3 9594.74 8686.58 84.56 90.47 79.80 84.72 87.88 91.68 0.94 0.94 1.04 1.01
4 3188.06 3068.35 109.21 114.26 108.00 114.60 118.12 123.66 0.99 1.00 1.08 1.08
5 900.21 941.75 127.92 136.72 128.68 134.01 147.27 141.45 1.01 0.98 1.15 1.03
6 136.87 298.75 144.54 139.28 141.93 144.62 156.63 158.95 0.98 1.04 1.08 1.14
7 21.04 73.31 129.45 129.46 125.35 115.32 135.78 132.20 0.97 0.89 1.05 1.02
8 7.87 18.90 176.98 154.07 161.11 152.48 176.05 160.04 0.91 0.99 0.99 1.04
9 15.02 20.31 106.11 206.69 160.39 148.29 159.40 160.72 1.51 0.72 1.50 0.78
10 7.18 1.16 10.89 124.34 63.58 102.14 68.12 96.06 5.84 0.82 6.26 0.77
11 0.05 4.59 151.46 209.79 109.63 113.56 105.98 96.27 0.72 0.54 0.70 0.46
12 0.10 0.71 111.86 210.32 74.69 118.83 101.45 128.47 0.67 0.56 0.91 0.61
Households by Drivers, Model Prediction vs. 2017 NHTS Data

Figure 37: Households by Drivers, Model Prediction vs. 2017 NHTS Data

Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Vehicles

Figure 38: Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Vehicles

Benchmarking Against 2009 Multimodal Model

The following tables and chart compare the performance of the 2009 and 2017 AADVMT Models by applying the 2009 model to the 2017 NHTS households.

Metro or Non-Metro Num HH NHTS2017 AADVMT/HH Model (2017) AADVMT/HH Model (2009) AADVMT/HH Ratio Model (2017)/NHTS2017 Ratio Model (2009)/NHTS2017
metro 71740 47.54 47.61 42.37 1 0.89
non_metro 56538 61.04 61.06 53.65 1 0.88
Scatterplot of 2009 Model Prediction vs. 2017 NHTS Data

Figure 39: Scatterplot of 2009 Model Prediction vs. 2017 NHTS Data

Scatterplot of 2009 Model Prediction vs. 2017 Model Prediction

Figure 40: Scatterplot of 2009 Model Prediction vs. 2017 Model Prediction

Budget Model Adjustment

Approach

The initial estimation of the model shows that, with the addition of the random variable factor drawn from the skewed normal distribution, the 2017 AADVMT model is able to replicate the mean AADVMT in the NHTS 2017, capture variation across the categories of explanatory variables, and capture household to household variability reasonably well.

However, the approach to estimating the model with reported AADVMT as the dependent variable is fundamentally different from what the model is tasked with predicting in the VE application. In the application, the AADVMT model is used to estimate household Dvmt based on the assumption that the amount of travel a household makes is not yet constrained by cost. Then a second model is applied in VE to reduce the Dvmt for households that exceed a certain household budget threshold for the maximum proportion of their income that they can spend on transportation costs.

Reported AADVMT in the survey is analogous with the budget adjusted Dvmt, i.e., it is the final actual travel that a household makes once any constraints such as spending have been considered in real life. Therefore, to be consistent with the model application approach and the use of the budget model, the dependent variable should be an estimated of household AADVMT that is unconstrained by costs.

In order to estimate a model using this approach, the model reported earlier in this document was applied in the Oregon implementation of VE-State and the proportional reduction in Dvmt between the application of the AADVMT model and the second of two iterations through the budget model was calculated for each household income category. This factor was then used to estimate a non-household budget constrained AADVMT for each household in 2017 NHTS. The AADVMT model was then re-estimated using these higher values of AADVMT.

Budget Adjusted Estimation and Prediction Results

The tables and charts below show the factors calculated from applying the model initially, the resulting changes in AADVMT by household income groups, the parameters of the re-estimated AADVMT model, and the results of applying the re-estimated model back to the 2017 NHTS data.

The appliations include the original VMT model estimated using the 2001 NHTS, the version of the AADVMT model estimated using the 2009 NHTS, and the initial version of the 2017 NHTS model.

HH Income Dvmt Dvmt (1st budget iteration) Dvmt (2nd budget iteration) Model Version Percent Increase in Dvmt Required
5000 28.35 15.08 14.65 MMNew 1.94
12000 35.28 26.54 26.01 MMNew 1.36
20000 39.07 31.38 30.88 MMNew 1.27
30000 42.70 36.13 35.65 MMNew 1.20
42000 46.58 40.93 40.47 MMNew 1.15
62000 51.38 46.43 46.03 MMNew 1.12
87000 56.00 51.79 51.39 MMNew 1.09
112000 59.88 56.29 55.91 MMNew 1.07
137000 63.29 60.02 59.68 MMNew 1.06
175000 67.28 64.40 64.08 MMNew 1.05
250000 68.74 66.48 66.26 MMNew 1.04
5000 21.33 14.76 14.39 MMOld 1.48
12000 33.02 27.25 26.74 MMOld 1.23
20000 38.54 33.11 32.58 MMOld 1.18
30000 44.14 39.15 38.62 MMOld 1.14
42000 49.78 45.16 44.63 MMOld 1.12
62000 56.37 52.20 51.70 MMOld 1.09
87000 62.90 59.18 58.72 MMOld 1.07
112000 68.18 64.87 64.43 MMOld 1.06
137000 72.92 69.80 69.38 MMOld 1.05
175000 78.44 75.56 75.18 MMOld 1.04
250000 81.38 79.07 78.80 MMOld 1.03
5000 25.55 16.19 15.78 Orig 1.62
12000 36.68 29.05 28.47 Orig 1.29
20000 41.76 34.78 34.18 Orig 1.22
30000 46.88 40.62 40.03 Orig 1.17
42000 52.15 46.48 45.89 Orig 1.14
62000 58.38 53.34 52.79 Orig 1.11
87000 64.52 60.15 59.64 Orig 1.08
112000 69.61 65.76 65.29 Orig 1.07
137000 74.16 70.65 70.18 Orig 1.06
175000 79.41 76.40 75.96 Orig 1.05
250000 86.10 83.16 82.86 Orig 1.04

The chart below shows the effect of appling the adjustments calculated from the 2017 NHTS model to the 2017 NHTS. The lower income categories are effected much more significantly than the higher income categories, flattening the impact of income. This is of course the expected outcome as lower income households are far more likely to hit a budget threshold even where they are assumed to spend higher overall proportions of their income on transportation,

Surveyed and Budget Adjusted AADVMT from the 2017 NHTS

Figure 41: Surveyed and Budget Adjusted AADVMT from the 2017 NHTS

The table below shows the re-estimated parameters of the AADVMT model, estimated using the budget adjusted AADVMT as the dependent variable.

Regression model summary for 2017 AADVMT Model with Budget Adjustment for Metro Areas and Non-Metro Areas
Dependent variable:
AADVMT
NONMETRO METRO
(1) (2)
Drivers 1.017*** (0.012) 1.077*** (0.009)
HhSize 0.106*** (0.011)
Workers 0.242*** (0.010) 0.145*** (0.009)
CENSUS_RNE -0.066*** (0.020) -0.090*** (0.017)
CENSUS_RS 0.099*** (0.015) 0.011 (0.015)
CENSUS_RW -0.217*** (0.018) -0.126*** (0.016)
FwyLaneMiPC 46.162** (18.093)
LogIncomeK 0.140*** (0.008) 0.017*** (0.006)
Age0to14 0.003 (0.013) 0.094*** (0.010)
Age65Plus -0.072*** (0.011) -0.107*** (0.011)
log1p(VehPerDriver) 4.565*** (0.026) 4.698*** (0.022)
LifeCycleCouple w/o children -0.030 (0.023) -0.062*** (0.016)
LifeCycleEmpty Nester -0.291*** (0.028) -0.494*** (0.022)
LifeCycleSingle -0.229*** (0.030) -0.440*** (0.019)
D1B -0.018*** (0.003) -0.002*** (0.0002)
D2A_EPHHM -0.308*** (0.031)
D1B:D2A_EPHHM 0.022*** (0.006)
D2A_WRKEMP -0.0002 (0.0002)
D3bpo4 -0.001*** (0.0001)
TranRevMiPC:D4c -0.047*** (0.005)
Constant -0.148*** (0.048) 0.238*** (0.033)
Observations 56,566 71,725
R2 0.630 0.689
Adjusted R2 0.630 0.689
Residual Std. Error 1.266 (df = 56549) 1.471 (df = 71707)
F Statistic 6,025.424*** (df = 16; 56549) 9,353.162*** (df = 17; 71707)
Note: p<0.1; p<0.05; p<0.01

When the model is applied back to the NHTS 2017 households, the results below show that, once the random variable simulation is complete, the results match the estimate of unconstrained household AADVMT. Reversing the factoring process results in an estimate of budget constrained household AADVMT that matches the original report AADVMT in the 2017 NHTS. It is likely that some calibration of the adjustment factors for each income categort that were used to factor the NHTS AADVMT for the budget adjusted estimation may be necessary. This is dependent on the response of the budget model during application testing.

Metro or Non-Metro Weighted HH NHTS2017 AADVMT/HH NHTS2017 Unconstrained AADVMT/HH Model AADVMT/HH Model Sim AADVMT/HH Model Budget Adj AADVMT/HH Ratio Model/ Uncon NHTS2017 Ratio Model Sim/ Uncon NHTS2017 Ratio Model Budget/ NHTS 2017
metro 81842.28 47.54 53.64 48.54 53.66 47.77 0.90 1 1.01
non_metro 46121.44 61.04 69.33 64.13 69.08 61.09 0.93 1 1.00